The U.S. government lifted export controls on Anthropic's top-tier artificial intelligence (AI) model "Claude Mythos."
Anthropic said on the 30th (local time) that it was notified the Commerce Department lifted, after 18 days, the export control guidance imposed on "Mythos 5" and "Fable 5."
Anthropic said, "We plan to resume service starting tomorrow," and added, "We thank users for their patience and everyone who worked to redistribute the model."
Mythos 5 was found to have expert-level capability to detect cybersecurity vulnerabilities, shocking the global security industry. Fable 5 is a model that adds safeguards to Mythos 5 to limit answers on sensitive topics that could be misused for hacking or weapons manufacturing.
On the 12th, the Commerce Department issued export control guidance barring foreign access to the two models, saying they posed a national security threat. The government was said to have issued the guidance out of concern that users could bypass the safeguards Anthropic applied to its models through a method known as "jailbreaking."
As a result, Anthropic, which lacked the capability to identify users' nationalities, blocked access to the two models not only for foreigners but also for Americans.
However, starting on the 26th, more than 100 domestic corporations and institutions approved by the U.S. government as trustworthy were allowed to use Mythos.